AB-testrun of Seer 2.3.0 finished. New engine in my list: All 7000 games played without any problems.
https://www.sp-cc.de
(Perhaps you have to clear your browsercache or reload the website)
SPCC: Testrun of Seer 2.3.0 finished
Moderators: hgm, Rebel, chrisw
-
- Posts: 2439
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl
-
- Posts: 533
- Joined: Sun Sep 06, 2020 4:40 am
- Full name: Connor McMonigle
Re: SPCC: Testrun of Seer 2.3.0 finished
Thanks for testing! That result looks very consistent with my testing against other engines. The listing seems to indicate the AVX binary was used. It shouldn't make a significant difference (definitely not worth testing again), but was there a reason for choosing the AVX binary over the AVX2 binary?pohl4711 wrote: ↑Fri Aug 20, 2021 3:43 pm AB-testrun of Seer 2.3.0 finished.
https://www.sp-cc.de
(Perhaps you have to clear your browsercache or reload the website)
-
- Posts: 2439
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl
Re: SPCC: Testrun of Seer 2.3.0 finished
Yes, the avx-Binary was little bit faster on my AMD Ryzen 12core CPU, than the avx2-Binary. But only some percent - no big deal. I measured all 3 binaries twice. Avx was the fastest.connor_mcmonigle wrote: ↑Fri Aug 20, 2021 3:49 pmThanks for testing! That result looks very consistent with my testing against other engines. The listing seems to indicate the AVX binary was used. It shouldn't make a significant difference (definitely not worth testing again), but was there a reason for choosing the AVX binary over the AVX2 binary?pohl4711 wrote: ↑Fri Aug 20, 2021 3:43 pm AB-testrun of Seer 2.3.0 finished.
https://www.sp-cc.de
(Perhaps you have to clear your browsercache or reload the website)
-
- Posts: 4468
- Joined: Fri Apr 21, 2006 4:19 pm
- Location: IASI - the historical capital of MOLDOVA
- Full name: SilvianR
Re: SPCC: Testrun of Seer 2.3.0 finished
-
- Posts: 2439
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl
Re: SPCC: Testrun of Seer 2.3.0 finished
So what? I have a total different CPU. I measured again: avx2 binary is around 5% slower than avx on my AMD Ryzen 3900 12core.
(5% speed difference is perhaps 2 Elo with my testing-conditions. Meaningless)
-
- Posts: 533
- Joined: Sun Sep 06, 2020 4:40 am
- Full name: Connor McMonigle
Re: SPCC: Testrun of Seer 2.3.0 finished
The hardware configuration is likely very relevant. It's possible that on Zen2 processors, FMA gets decoded into the same micro ops as a MUL followed by an ADD which would result in the behavior Pohl is observing.
Regardless, just one comparison of nps is insufficient as the variance is likely significant. Furthermore, just testing nps on the start position is also flawed. A more accurate picture is obtained by running bench several times and averaging across all trials. To compare NPS, I use this script: https://github.com/TerjeKir/EngineTests ... speedup.py. In any case, the percent nps difference is so tiny that it won't meaningfully influence the performance (certainly within error of a 100 game test, lol).
Regardless, just one comparison of nps is insufficient as the variance is likely significant. Furthermore, just testing nps on the start position is also flawed. A more accurate picture is obtained by running bench several times and averaging across all trials. To compare NPS, I use this script: https://github.com/TerjeKir/EngineTests ... speedup.py. In any case, the percent nps difference is so tiny that it won't meaningfully influence the performance (certainly within error of a 100 game test, lol).
-
- Posts: 4468
- Joined: Fri Apr 21, 2006 4:19 pm
- Location: IASI - the historical capital of MOLDOVA
- Full name: SilvianR